massively parallel molecular-continuum simulations with ... · viewed, and aspects for massively...
TRANSCRIPT
References[1] P. Neumann and N. Tchipev, A Coupling Tool for Parallel Molecular Dynamics–Continuum Simulations, Proceedings of the International Symposium on Parallel and Distributed Computing, 2012[2] R. Delgado-Buscalioni, P.V. Coveney. USHER: An algorithm for particle insertion in dense fluids. J. Chem. Phys. 119 (2): 978–987 (2003).
[3] A. Dupuis, E.M. Kotsalis, P. Koumoutsakos. Coupling lattice Boltzmann and molecular dynamics models for dense fluids. Phys. Rev. E 75, 046704 (2007).
1 Department of Computer Science, Technische Universität München, Boltzmannstr. 3, D-85748 Garching, Germany, [email protected] 2 Department of Applied Physics, Technische Universiteit Eindhoven, P.O. Box 513, 5600 MB Eindhoven, The Netherlands, [email protected]
'
&
$
%
Outlook: Nanofilters
• Flow between two reservoirs through amolecular filter• Flow simulation in reservoirs: LB• Simulation of flow through filter: MD• Fully three-dimensional steady-state cou-
pling based on [3]
'
&
$
%
MaMiCo: Macro-Micro-Coupling Tool
ParticleInsertion
MomentumInsertion
MacroscopicCellServiceMacroscopicCellService
MomentumController
KineticEnergyController
MoleculeWrapper
MoleculeIterator
MDSolverInterfaceMacroscopicSolverInterface
MDSimulation ContinuumSimulation
TransferStrategy
implements
utilises
utilises
sendsdata to
receivesdata from/
sendsdata to
implements
utilises utilises
MaMiCo
User/Programmer
uses/ provides newimplementation of
•C++-development for spatiotemporal hybridmolecular–continuum simulations [1]• Support of 2D/ 3D hybrid scenarios•Macroscopic cells within MaMiCo are used to
– map macroscopic quantities between themacroscopic solver and MaMiCo
– map macro-/microscopic quantities betweenthe MD solver and MaMiCo
• Particle insertion/ deletion [2] encapsulatedwithin MaMiCo
Macroscopic Solver
MaMiCo
Molecular Dynamics
Particle Insertion/Deletion
Momentum
MassMomentum
MassMomentum
MassMomentum
'
&
$
%
Macroscopic Cell Exchange•MPI-based exchange of macroscopic cell information between MD and LB• Exchange (mostly) hidden from user• Example: LB on one process, MD on four processes
LB
MD
LB
MD
MacroscopicCellService→ Send to MD
Start: traverse LB cells. Step 1: track cells that should be sentto MD
LB
MD
MacroscopicCellService→ Receive from MD
LB
MD
MacroscopicCellService→ Setup Send-Recv
Step 2: track cells that should be re-ceived from MD
Step 3: setup send-recv topologywithin MacroscopicCellService
'
&
$
%
Parallel Results
MD on rank 0
LB domain andembedded MD
•Channel flow simulation with MD region embedded in the center• LB solver: sequential, 54× 54× 54 cells, BGK•MD solver: parallel, 1 000 000 single-centered LJ molecules• LB→ MD: Velocity relaxation in (MD) boundary cells, retaining mass in outermost cells
using USHER•MD→ LB: Only map velocities and incorporate them into particle distribution functions
Proc. Shaheen HuygensMD-LB MD MD-LB MD
1 1.0 1.0 1.0 1.08 7.2 7.3 7.3 6.9
64 46.0 49.6 45.7 45.3512 244.0 321.0 235.5 249.7
1728 484.4 814.5 456.7 494.7
Strong scaling for channel flow scenario. MD-LB: coupled simulation.MD: reference MD simulation for same MD domain without coupling.
1 10 100 1000 10000
0.7
0.78
0.86
0.94
1.02
Huygens t0Huygens t1Huygens t2Shaheen t0Shaheen t1Shaheen t2
# cores
Effi
cien
cy
Weak scaling for parallel USHER-based particle insertion [1]
'
&
$
%
Overview
Efficient implementations of hybrid molecular-continuum flow solvers are required toallow for fast and massively parallel simulations of large complex systems. Several cou-pling strategies have been proposed over the last years for 2D/ 3D, time-dependent/steady-state or compressible/ incompressible scenarios. Despite their different applica-tion areas, most of these schemes comprise the same or similar building blocks. Still, tothe authors’ knowledge, no common implementation of these building blocks is availableyet. In this contribution, the Macro-Micro-Coupling Tool is presented which is meant tosupport developers in coupling mesh-based methods with molecular dynamics. It iswritten in C++ and supports two- and three-dimensional scenarios. Its design is re-viewed, and aspects for massively parallel coupled scenarios are addressed. Scalingresults are presented for a hybrid simulation which couples a molecular dynamics codeto the Lattice Boltzmann application of the Peano framework.
Philipp Neumann1 Jens Harting2
Massively Parallel Molecular-Continuum Simulationswith the Macro-Micro-Coupling Tool